add_ebb_estimate: Perform empirical Bayes shrinkage on a data frame

In dgrtwo/ebbinom: Empirical Bayes on binomial data

Description

Add columns to a data frame representing empirical Bayes shrinkage towards an estimated beta prior. This calls the ebb_fit_prior function to fit the prior, then adds the columns using augment.ebb_prior. It is thus a useful wrapper when you're not interested in the prior itself, but only in performing shrinkage on data.

Usage

add_ebb_estimate(tbl, x, n, cred_level = 0.95, prior_subset = TRUE, ...)

add_ebb_estimate_(tbl, x, n, prior_subset = TRUE, cred_level = 0.05, ...)

Arguments

tbl	A table.
x	Column containing number of successes.
n	Column containing totals.
cred_level	Level of credible interval to compute. If NULL, do not compute intervals.
prior_subset	An expression evaluating to a logical vector indicating which values should be used for computing the prior.
...	Extra arguments passed on to ebb_fit_prior, such as method.

Value

The original table, with several columns added based on empirical Bayes shrinkage:

.alpha1	Posterior alpha (shape1) parameter
.beta1	Posterior beta (shape2) parameter
.fitted	Posterior shrunken estimate
.raw	Estimate without shrinkage (success / total)
.low	Lower bound of credible interval
.high	Upper bound of credible interval

See Also

ebb_prior_tidiers

Examples

library(dplyr)
set.seed(2017)

# simulate 200 random examples from a beta-binomial
obs <- 200
dat <- data_frame(prob = rbeta(obs, 10, 50),
                  total = round(rlnorm(obs, 4, 2)) + 1,
                  x = rbinom(obs, total, prob))

result <- add_ebb_estimate(dat, x, total)
result

# visualize the shrinkage towards the prior mean
library(ggplot2)
ggplot(result, aes(.raw, .fitted, color = log10(total))) +
  geom_point() +
  geom_abline(color = "red")

dgrtwo/ebbinom documentation built on May 15, 2019, 7:23 a.m.